mofa-org/mofa-skills/mofa-notebook-video-overview

Notebook Video Overview

Use this skill when the user wants a video overview of notebook sources. It calls a configured LLM with labelled notebook chunks, validates returned citation chunk IDs, writes grounded script / scene / asset files, then renders a short MP4 with Veo 3.1 by default.

Call video_overview_generate once. The result includes overview.mp4, veo-prompt.txt, veo-operation.json, script.md, scene-plan.json, asset-brief.md, and handoff.md under notebook-outputs/video-overviews/. Set render_video to false only when the user explicitly asks for planning files without spending video-generation quota.

Optional inputs include title, style, duration_minutes, source_ids, language, provider, model, video_model, video_duration_seconds, video_aspect_ratio, video_resolution, video_provider, and video_output_gcs_uri.

Runtime

The shared notebook runtime is bundled inside the skill, so standalone installations do not need an extra notebook_common directory.

Model Configuration

The planning step uses the shared notebook LLM configuration: GEMINI_API_KEY, OPENAI_API_KEY, or Vertex via GOOGLE_APPLICATION_CREDENTIALS / VERTEX_SA_JSON. Optional defaults are MOFA_NOTEBOOK_PROVIDER and MOFA_NOTEBOOK_MODEL; MOFA_DATA_TABLE_PROVIDER and MOFA_DATA_TABLE_MODEL remain accepted for compatibility.

The rendering step supports Gemini API Veo or Vertex Veo. Gemini API rendering uses GEMINI_API_KEY and defaults to veo-3.1-generate-preview. Vertex rendering uses GOOGLE_APPLICATION_CREDENTIALS, VERTEX_SA_JSON, VERTEX_ACCESS_TOKEN, or GOOGLE_OAUTH_ACCESS_TOKEN; it also needs GOOGLE_CLOUD_PROJECT unless the service account JSON includes project_id, and defaults to GOOGLE_CLOUD_LOCATION=us-central1 with veo-3.1-generate-001. Override with video_provider, MOFA_NOTEBOOK_VIDEO_PROVIDER, MOFA_VEO_PROVIDER, video_model, MOFA_NOTEBOOK_VIDEO_MODEL, or MOFA_VEO_MODEL.

For Vertex, video_output_gcs_uri, MOFA_VEO_OUTPUT_GCS_URI, or VERTEX_OUTPUT_GCS_URI may point to a gs://... output prefix. The configured credential must be able to read the generated MP4 so the skill can attach overview.mp4 locally.